Sergei, Thank you for your time and responses. On Tue, Feb 14, 2017 at 3:11 PM Sergei Golubchik <serg@mariadb.org> wrote:
Hi, Seth!
It's not what SQL standard specifies. It says,
<...> whenever a datetime value without time zone is to be implicitly derived from one with, SQL assumes the value with time zone to be UTC, adds the time zone displacement to it to give local time, and the result, without any time zone displacement, is local.
This doesn't make much sense to be, really, because it totally ignores the session time zone. On the other hand, if you'd convert the UTC value to be in the session time zone, you'd totally ignore the displacement value. I don't see any behavior being better than the other, so I'd stick to the standard by default.
In order to convert from RFC3339 timezone to session timezone, we will need to return the timezone from the str_to_datetime function. As mentioned there is no timezone member of the MY_TIME struct. I believe a change to the MY_TIME struct would be the most beneficial and cleanest approach. This would allow for handling RFC3339 and future additions, such as a datetime with time zone field, similar to the ansi timestamp with time zone. The struct is used in a several ways, including packing into longlong, all of those cases would need to be handled. This would be a much larger and more involved patchset.
To support TIMESTAMP WITH TIME ZONE and TIME WITH TIME ZONE literals (literals only, not expressions or fields or variables), it would be enough to parse them in str_to_datetime(), there is no need to return the time zone back to the caller in MY_TIME.
In adding a gmt_offset variable, and boolean flag to MY_TIME struct it will break abi compatibility. What are the standards around abi compatibility changes? Would adding all relevant changes behind #ifdef be sufficient? In that case, at compile time a user could choose to maintain existing abi compatibility, or choose to enable TIME(STAMP) with TIME ZONE literal support?
With any approach, the timezone could then be handled outside of my_time.c, where we have THD and the timezone class, which can easily handle the conversion.
There is no need to handle time zone outside of my_time.c if all we want is TIMESTAMP WITH TIME ZONE literals with the standard conversion behavior. It can be implemented completely inside str_to_datetime.
Regards, Sergei Chief Architect MariaDB and security@mariadb.org
Thank you, Seth Shelnutt